Interview with Ai2's Former RLHF Lead: Open Models Rise and RSI Myths

ziv_ravid · x · 2026-08-09

This podcast episode features Nathan Lambert, who previously led post-training at Ai2 and authored the book on RLHF. The discussion centers on the rapid improvement of open-source models and their implications for the next 2-5 years.

Lambert shares several key insights: he was surprised by how good open models have gotten recently; he believes the hype around recursive self-improvement (RSI) is currently overblown; and he predicts that Anthropic's famously open internal culture might crack under the pressures of an IPO.

Original post →

More from Companies & People

Companies & People channel →